THU-DSP-LAB
/
llvm-project
mirror of https://github.com/THU-DSP-LAB/llvm-project.git
2
0
Fork 1
Code Issues Packages Projects Releases Wiki Activity
llvm-project/llvm/test/CodeGen/AMDGPU/fcanonicalize.f16.ll

2234 lines
82 KiB
LLVM
Raw Permalink Blame History

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=VI %s
; RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX9 %s
; RUN: llc -march=amdgcn -mcpu=kaveri -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=CI %s
declare half @llvm.fabs.f16(half) #0
declare half @llvm.canonicalize.f16(half) #0
declare <2 x half> @llvm.fabs.v2f16(<2 x half>) #0
declare <2 x half> @llvm.canonicalize.v2f16(<2 x half>) #0
declare <3 x half> @llvm.canonicalize.v3f16(<3 x half>) #0
declare <4 x half> @llvm.canonicalize.v4f16(<4 x half>) #0
declare i32 @llvm.amdgcn.workitem.id.x() #0
define amdgpu_kernel void @test_fold_canonicalize_undef_value_f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: test_fold_canonicalize_undef_value_f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_mov_b32_e32 v2, 0
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_short v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: test_fold_canonicalize_undef_value_f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_short v0, v0, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: test_fold_canonicalize_undef_value_f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: v_mov_b32_e32 v0, 0
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%canonicalized = call half @llvm.canonicalize.f16(half undef)
store half %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @v_test_canonicalize_var_f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: v_test_canonicalize_var_f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_load_ushort v0, v[0:1]
; VI-NEXT: s_waitcnt vmcnt(0)
; VI-NEXT: v_max_f16_e32 v0, v0, v0
; VI-NEXT: flat_store_short v[0:1], v0
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: v_test_canonicalize_var_f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_load_ushort v0, v0, s[0:1]
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_max_f16_e32 v0, v0, v0
; GFX9-NEXT: global_store_short v[0:1], v0, off
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: v_test_canonicalize_var_f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_load_ushort v0, off, s[0:3], 0
; CI-NEXT: s_waitcnt vmcnt(0)
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
; CI-NEXT: v_mul_f32_e32 v0, 1.0, v0
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%val = load half, ptr addrspace(1) %out
%canonicalized = call half @llvm.canonicalize.f16(half %val)
store half %canonicalized, ptr addrspace(1) undef
ret void
}
define amdgpu_kernel void @s_test_canonicalize_var_f16(ptr addrspace(1) %out, i16 zeroext %val.arg) #1 {
; VI-LABEL: s_test_canonicalize_var_f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dword s2, s[0:1], 0x2c
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_max_f16_e64 v2, s2, s2
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_short v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: s_test_canonicalize_var_f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c
; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: v_max_f16_e64 v1, s4, s4
; GFX9-NEXT: global_store_short v0, v1, s[2:3]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: s_test_canonicalize_var_f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dword s2, s[0:1], 0xb
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: v_cvt_f32_f16_e32 v0, s2
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: v_mul_f32_e32 v0, 1.0, v0
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%val = bitcast i16 %val.arg to half
%canonicalized = call half @llvm.canonicalize.f16(half %val)
store half %canonicalized, ptr addrspace(1) %out
ret void
}
define <2 x half> @v_test_canonicalize_build_vector_v2f16(half %lo, half %hi) #1 {
; VI-LABEL: v_test_canonicalize_build_vector_v2f16:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_max_f16_sdwa v1, v1, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_max_f16_e32 v0, v0, v0
; VI-NEXT: v_or_b32_e32 v0, v0, v1
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: v_test_canonicalize_build_vector_v2f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: s_mov_b32 s4, 0x5040100
; GFX9-NEXT: v_perm_b32 v0, v1, v0, s4
; GFX9-NEXT: v_pk_max_f16 v0, v0, v0
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; CI-LABEL: v_test_canonicalize_build_vector_v2f16:
; CI: ; %bb.0:
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
; CI-NEXT: v_cvt_f16_f32_e32 v1, v1
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
; CI-NEXT: v_cvt_f32_f16_e32 v1, v1
; CI-NEXT: s_setpc_b64 s[30:31]
%ins0 = insertelement <2 x half> undef, half %lo, i32 0
%ins1 = insertelement <2 x half> %ins0, half %hi, i32 1
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %ins1)
ret <2 x half> %canonicalized
}
define amdgpu_kernel void @v_test_canonicalize_fabs_var_f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: v_test_canonicalize_fabs_var_f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_load_ushort v2, v[0:1]
; VI-NEXT: s_waitcnt vmcnt(0)
; VI-NEXT: v_max_f16_e64 v2, |v2|, |v2|
; VI-NEXT: flat_store_short v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: v_test_canonicalize_fabs_var_f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_load_ushort v1, v0, s[0:1]
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_max_f16_e64 v1, |v1|, |v1|
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: v_test_canonicalize_fabs_var_f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_load_ushort v0, off, s[0:3], 0
; CI-NEXT: s_waitcnt vmcnt(0)
; CI-NEXT: v_cvt_f32_f16_e64 v0, |v0|
; CI-NEXT: v_mul_f32_e32 v0, 1.0, v0
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%val = load half, ptr addrspace(1) %out
%val.fabs = call half @llvm.fabs.f16(half %val)
%canonicalized = call half @llvm.canonicalize.f16(half %val.fabs)
store half %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @v_test_canonicalize_fneg_fabs_var_f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: v_test_canonicalize_fneg_fabs_var_f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_load_ushort v2, v[0:1]
; VI-NEXT: s_waitcnt vmcnt(0)
; VI-NEXT: v_max_f16_e64 v2, -|v2|, -|v2|
; VI-NEXT: flat_store_short v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: v_test_canonicalize_fneg_fabs_var_f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_load_ushort v1, v0, s[0:1]
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_max_f16_e64 v1, -|v1|, -|v1|
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: v_test_canonicalize_fneg_fabs_var_f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_load_ushort v0, off, s[0:3], 0
; CI-NEXT: s_waitcnt vmcnt(0)
; CI-NEXT: v_cvt_f32_f16_e64 v0, -|v0|
; CI-NEXT: v_mul_f32_e32 v0, 1.0, v0
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%val = load half, ptr addrspace(1) %out
%val.fabs = call half @llvm.fabs.f16(half %val)
%val.fabs.fneg = fneg half %val.fabs
%canonicalized = call half @llvm.canonicalize.f16(half %val.fabs.fneg)
store half %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @v_test_canonicalize_fneg_var_f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: v_test_canonicalize_fneg_var_f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_load_ushort v2, v[0:1]
; VI-NEXT: s_waitcnt vmcnt(0)
; VI-NEXT: v_max_f16_e64 v2, -v2, -v2
; VI-NEXT: flat_store_short v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: v_test_canonicalize_fneg_var_f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_load_ushort v1, v0, s[0:1]
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_max_f16_e64 v1, -v1, -v1
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: v_test_canonicalize_fneg_var_f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_load_ushort v0, off, s[0:3], 0
; CI-NEXT: s_waitcnt vmcnt(0)
; CI-NEXT: v_cvt_f32_f16_e64 v0, -v0
; CI-NEXT: v_mul_f32_e32 v0, 1.0, v0
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%val = load half, ptr addrspace(1) %out
%val.fneg = fneg half %val
%canonicalized = call half @llvm.canonicalize.f16(half %val.fneg)
store half %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @v_test_no_denormals_canonicalize_fneg_var_f16(ptr addrspace(1) %out) #2 {
; VI-LABEL: v_test_no_denormals_canonicalize_fneg_var_f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_load_ushort v2, v[0:1]
; VI-NEXT: s_waitcnt vmcnt(0)
; VI-NEXT: v_mul_f16_e32 v2, -1.0, v2
; VI-NEXT: flat_store_short v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: v_test_no_denormals_canonicalize_fneg_var_f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_load_ushort v1, v0, s[0:1]
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_max_f16_e64 v1, -v1, -v1
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: v_test_no_denormals_canonicalize_fneg_var_f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_load_ushort v0, off, s[0:3], 0
; CI-NEXT: s_waitcnt vmcnt(0)
; CI-NEXT: v_cvt_f32_f16_e64 v0, -v0
; CI-NEXT: v_mul_f32_e32 v0, 1.0, v0
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%val = load half, ptr addrspace(1) %out
%val.fneg = fneg half %val
%canonicalized = call half @llvm.canonicalize.f16(half %val.fneg)
store half %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @v_test_no_denormals_canonicalize_fneg_fabs_var_f16(ptr addrspace(1) %out) #2 {
; VI-LABEL: v_test_no_denormals_canonicalize_fneg_fabs_var_f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_load_ushort v2, v[0:1]
; VI-NEXT: s_waitcnt vmcnt(0)
; VI-NEXT: v_mul_f16_e64 v2, -1.0, |v2|
; VI-NEXT: flat_store_short v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: v_test_no_denormals_canonicalize_fneg_fabs_var_f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_load_ushort v1, v0, s[0:1]
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_max_f16_e64 v1, -|v1|, -|v1|
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: v_test_no_denormals_canonicalize_fneg_fabs_var_f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_load_ushort v0, off, s[0:3], 0
; CI-NEXT: s_waitcnt vmcnt(0)
; CI-NEXT: v_cvt_f32_f16_e64 v0, -|v0|
; CI-NEXT: v_mul_f32_e32 v0, 1.0, v0
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%val = load half, ptr addrspace(1) %out
%val.fabs = call half @llvm.fabs.f16(half %val)
%val.fabs.fneg = fneg half %val.fabs
%canonicalized = call half @llvm.canonicalize.f16(half %val.fabs.fneg)
store half %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @test_fold_canonicalize_p0_f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: test_fold_canonicalize_p0_f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_mov_b32_e32 v2, 0
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_short v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: test_fold_canonicalize_p0_f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_short v0, v0, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: test_fold_canonicalize_p0_f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: v_mov_b32_e32 v0, 0
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%canonicalized = call half @llvm.canonicalize.f16(half 0.0)
store half %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @test_fold_canonicalize_n0_f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: test_fold_canonicalize_n0_f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_mov_b32_e32 v2, 0xffff8000
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_short v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: test_fold_canonicalize_n0_f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: v_mov_b32_e32 v1, 0xffff8000
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: test_fold_canonicalize_n0_f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: v_mov_b32_e32 v0, 0x8000
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%canonicalized = call half @llvm.canonicalize.f16(half -0.0)
store half %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @test_fold_canonicalize_p1_f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: test_fold_canonicalize_p1_f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_mov_b32_e32 v2, 0x3c00
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_short v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: test_fold_canonicalize_p1_f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: v_mov_b32_e32 v1, 0x3c00
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: test_fold_canonicalize_p1_f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: v_mov_b32_e32 v0, 0x3c00
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%canonicalized = call half @llvm.canonicalize.f16(half 1.0)
store half %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @test_fold_canonicalize_n1_f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: test_fold_canonicalize_n1_f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_mov_b32_e32 v2, 0xffffbc00
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_short v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: test_fold_canonicalize_n1_f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: v_mov_b32_e32 v1, 0xffffbc00
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: test_fold_canonicalize_n1_f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: v_mov_b32_e32 v0, 0xbc00
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%canonicalized = call half @llvm.canonicalize.f16(half -1.0)
store half %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @test_fold_canonicalize_literal_f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: test_fold_canonicalize_literal_f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_mov_b32_e32 v2, 0x4c00
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_short v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: test_fold_canonicalize_literal_f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: v_mov_b32_e32 v1, 0x4c00
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: test_fold_canonicalize_literal_f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: v_mov_b32_e32 v0, 0x4c00
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%canonicalized = call half @llvm.canonicalize.f16(half 16.0)
store half %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @test_default_denormals_fold_canonicalize_denormal0_f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: test_default_denormals_fold_canonicalize_denormal0_f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_mov_b32_e32 v2, 0x3ff
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_short v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: test_default_denormals_fold_canonicalize_denormal0_f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: v_mov_b32_e32 v1, 0x3ff
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: test_default_denormals_fold_canonicalize_denormal0_f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: v_mov_b32_e32 v0, 0x3ff
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%canonicalized = call half @llvm.canonicalize.f16(half 0xH03FF)
store half %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @test_denormals_fold_canonicalize_denormal0_f16(ptr addrspace(1) %out) #3 {
; VI-LABEL: test_denormals_fold_canonicalize_denormal0_f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_mov_b32_e32 v2, 0x3ff
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_short v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: test_denormals_fold_canonicalize_denormal0_f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: v_mov_b32_e32 v1, 0x3ff
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: test_denormals_fold_canonicalize_denormal0_f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: v_mov_b32_e32 v0, 0x3ff
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%canonicalized = call half @llvm.canonicalize.f16(half 0xH03FF)
store half %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @test_default_denormals_fold_canonicalize_denormal1_f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: test_default_denormals_fold_canonicalize_denormal1_f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_mov_b32_e32 v2, 0xffff83ff
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_short v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: test_default_denormals_fold_canonicalize_denormal1_f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: v_mov_b32_e32 v1, 0xffff83ff
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: test_default_denormals_fold_canonicalize_denormal1_f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: v_mov_b32_e32 v0, 0x83ff
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%canonicalized = call half @llvm.canonicalize.f16(half 0xH83FF)
store half %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @test_denormals_fold_canonicalize_denormal1_f16(ptr addrspace(1) %out) #3 {
; VI-LABEL: test_denormals_fold_canonicalize_denormal1_f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_mov_b32_e32 v2, 0xffff83ff
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_short v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: test_denormals_fold_canonicalize_denormal1_f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: v_mov_b32_e32 v1, 0xffff83ff
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: test_denormals_fold_canonicalize_denormal1_f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: v_mov_b32_e32 v0, 0x83ff
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%canonicalized = call half @llvm.canonicalize.f16(half 0xH83FF)
store half %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @test_fold_canonicalize_qnan_f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: test_fold_canonicalize_qnan_f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_mov_b32_e32 v2, 0x7c00
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_short v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: test_fold_canonicalize_qnan_f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: v_mov_b32_e32 v1, 0x7c00
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: test_fold_canonicalize_qnan_f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: v_mov_b32_e32 v0, 0x7c00
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%canonicalized = call half @llvm.canonicalize.f16(half 0xH7C00)
store half %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @test_fold_canonicalize_qnan_value_neg1_f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: test_fold_canonicalize_qnan_value_neg1_f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_mov_b32_e32 v2, 0x7e00
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_short v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: test_fold_canonicalize_qnan_value_neg1_f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: v_mov_b32_e32 v1, 0x7e00
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: test_fold_canonicalize_qnan_value_neg1_f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: v_mov_b32_e32 v0, 0x7e00
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%canonicalized = call half @llvm.canonicalize.f16(half bitcast (i16 -1 to half))
store half %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @test_fold_canonicalize_qnan_value_neg2_f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: test_fold_canonicalize_qnan_value_neg2_f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_mov_b32_e32 v2, 0x7e00
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_short v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: test_fold_canonicalize_qnan_value_neg2_f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: v_mov_b32_e32 v1, 0x7e00
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: test_fold_canonicalize_qnan_value_neg2_f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: v_mov_b32_e32 v0, 0x7e00
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%canonicalized = call half @llvm.canonicalize.f16(half bitcast (i16 -2 to half))
store half %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @test_fold_canonicalize_snan0_value_f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: test_fold_canonicalize_snan0_value_f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_mov_b32_e32 v2, 0x7e00
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_short v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: test_fold_canonicalize_snan0_value_f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: v_mov_b32_e32 v1, 0x7e00
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: test_fold_canonicalize_snan0_value_f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: v_mov_b32_e32 v0, 0x7e00
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%canonicalized = call half @llvm.canonicalize.f16(half 0xH7C01)
store half %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @test_fold_canonicalize_snan1_value_f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: test_fold_canonicalize_snan1_value_f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_mov_b32_e32 v2, 0x7e00
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_short v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: test_fold_canonicalize_snan1_value_f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: v_mov_b32_e32 v1, 0x7e00
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: test_fold_canonicalize_snan1_value_f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: v_mov_b32_e32 v0, 0x7e00
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%canonicalized = call half @llvm.canonicalize.f16(half 0xH7DFF)
store half %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @test_fold_canonicalize_snan2_value_f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: test_fold_canonicalize_snan2_value_f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_mov_b32_e32 v2, 0x7e00
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_short v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: test_fold_canonicalize_snan2_value_f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: v_mov_b32_e32 v1, 0x7e00
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: test_fold_canonicalize_snan2_value_f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: v_mov_b32_e32 v0, 0x7e00
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%canonicalized = call half @llvm.canonicalize.f16(half 0xHFDFF)
store half %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @test_fold_canonicalize_snan3_value_f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: test_fold_canonicalize_snan3_value_f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_mov_b32_e32 v2, 0x7e00
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_short v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: test_fold_canonicalize_snan3_value_f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: v_mov_b32_e32 v1, 0x7e00
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_short v0, v1, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: test_fold_canonicalize_snan3_value_f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: v_mov_b32_e32 v0, 0x7e00
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_store_short v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%canonicalized = call half @llvm.canonicalize.f16(half 0xHFC01)
store half %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @v_test_canonicalize_var_v2f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: v_test_canonicalize_var_v2f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_lshlrev_b32_e32 v0, 2, v0
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: v_add_u32_e32 v0, vcc, s0, v0
; VI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT: flat_load_dword v0, v[0:1]
; VI-NEXT: s_waitcnt vmcnt(0)
; VI-NEXT: v_max_f16_sdwa v1, v0, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
; VI-NEXT: v_max_f16_e32 v0, v0, v0
; VI-NEXT: v_or_b32_e32 v2, v0, v1
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_dword v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: v_test_canonicalize_var_v2f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0
; GFX9-NEXT: v_mov_b32_e32 v1, 0
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_load_dword v0, v0, s[0:1]
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_pk_max_f16 v0, v0, v0
; GFX9-NEXT: global_store_dword v1, v0, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: v_test_canonicalize_var_v2f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s6, 0
; CI-NEXT: s_mov_b32 s7, s3
; CI-NEXT: v_lshlrev_b32_e32 v0, 2, v0
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: s_mov_b64 s[4:5], s[0:1]
; CI-NEXT: v_mov_b32_e32 v1, 0
; CI-NEXT: buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: s_waitcnt vmcnt(0)
; CI-NEXT: v_lshrrev_b32_e32 v1, 16, v0
; CI-NEXT: v_cvt_f32_f16_e32 v1, v1
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
; CI-NEXT: v_mul_f32_e32 v1, 1.0, v1
; CI-NEXT: v_cvt_f16_f32_e32 v1, v1
; CI-NEXT: v_mul_f32_e32 v0, 1.0, v0
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
; CI-NEXT: v_lshlrev_b32_e32 v1, 16, v1
; CI-NEXT: v_or_b32_e32 v0, v0, v1
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep = getelementptr <2 x half>, ptr addrspace(1) %out, i32 %tid
%val = load <2 x half>, ptr addrspace(1) %gep
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %val)
store <2 x half> %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @v_test_canonicalize_fabs_var_v2f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: v_test_canonicalize_fabs_var_v2f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_lshlrev_b32_e32 v0, 2, v0
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: v_add_u32_e32 v0, vcc, s0, v0
; VI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT: flat_load_dword v0, v[0:1]
; VI-NEXT: s_waitcnt vmcnt(0)
; VI-NEXT: v_max_f16_sdwa v1, |v0|, |v0| dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
; VI-NEXT: v_max_f16_e64 v0, |v0|, |v0|
; VI-NEXT: v_or_b32_e32 v2, v0, v1
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_dword v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: v_test_canonicalize_fabs_var_v2f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0
; GFX9-NEXT: v_mov_b32_e32 v1, 0
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_load_dword v0, v0, s[0:1]
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v0
; GFX9-NEXT: v_pk_max_f16 v0, v0, v0
; GFX9-NEXT: global_store_dword v1, v0, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: v_test_canonicalize_fabs_var_v2f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s6, 0
; CI-NEXT: s_mov_b32 s7, s3
; CI-NEXT: v_lshlrev_b32_e32 v0, 2, v0
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: s_mov_b64 s[4:5], s[0:1]
; CI-NEXT: v_mov_b32_e32 v1, 0
; CI-NEXT: buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: s_waitcnt vmcnt(0)
; CI-NEXT: v_lshrrev_b32_e32 v1, 16, v0
; CI-NEXT: v_cvt_f32_f16_e64 v1, |v1|
; CI-NEXT: v_cvt_f32_f16_e64 v0, |v0|
; CI-NEXT: v_mul_f32_e32 v1, 1.0, v1
; CI-NEXT: v_cvt_f16_f32_e32 v1, v1
; CI-NEXT: v_mul_f32_e32 v0, 1.0, v0
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
; CI-NEXT: v_lshlrev_b32_e32 v1, 16, v1
; CI-NEXT: v_or_b32_e32 v0, v0, v1
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep = getelementptr <2 x half>, ptr addrspace(1) %out, i32 %tid
%val = load <2 x half>, ptr addrspace(1) %gep
%val.fabs = call <2 x half> @llvm.fabs.v2f16(<2 x half> %val)
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %val.fabs)
store <2 x half> %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @v_test_canonicalize_fneg_fabs_var_v2f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: v_test_canonicalize_fneg_fabs_var_v2f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_lshlrev_b32_e32 v0, 2, v0
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: v_add_u32_e32 v0, vcc, s0, v0
; VI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT: flat_load_dword v0, v[0:1]
; VI-NEXT: s_waitcnt vmcnt(0)
; VI-NEXT: v_max_f16_sdwa v1, -|v0|, -|v0| dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
; VI-NEXT: v_max_f16_e64 v0, -|v0|, -|v0|
; VI-NEXT: v_or_b32_e32 v2, v0, v1
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_dword v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: v_test_canonicalize_fneg_fabs_var_v2f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0
; GFX9-NEXT: v_mov_b32_e32 v1, 0
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_load_dword v0, v0, s[0:1]
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v0
; GFX9-NEXT: v_pk_max_f16 v0, v0, v0 neg_lo:[1,1] neg_hi:[1,1]
; GFX9-NEXT: global_store_dword v1, v0, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: v_test_canonicalize_fneg_fabs_var_v2f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s6, 0
; CI-NEXT: s_mov_b32 s7, s3
; CI-NEXT: v_lshlrev_b32_e32 v0, 2, v0
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: s_mov_b64 s[4:5], s[0:1]
; CI-NEXT: v_mov_b32_e32 v1, 0
; CI-NEXT: buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: s_waitcnt vmcnt(0)
; CI-NEXT: v_or_b32_e32 v0, 0x80008000, v0
; CI-NEXT: v_lshrrev_b32_e32 v1, 16, v0
; CI-NEXT: v_cvt_f32_f16_e32 v1, v1
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
; CI-NEXT: v_mul_f32_e32 v1, 1.0, v1
; CI-NEXT: v_cvt_f16_f32_e32 v1, v1
; CI-NEXT: v_mul_f32_e32 v0, 1.0, v0
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
; CI-NEXT: v_lshlrev_b32_e32 v1, 16, v1
; CI-NEXT: v_or_b32_e32 v0, v0, v1
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep = getelementptr <2 x half>, ptr addrspace(1) %out, i32 %tid
%val = load <2 x half>, ptr addrspace(1) %gep
%val.fabs = call <2 x half> @llvm.fabs.v2f16(<2 x half> %val)
%val.fabs.fneg = fneg <2 x half> %val.fabs
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %val.fabs.fneg)
store <2 x half> %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @v_test_canonicalize_fneg_var_v2f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: v_test_canonicalize_fneg_var_v2f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_lshlrev_b32_e32 v0, 2, v0
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: v_add_u32_e32 v0, vcc, s0, v0
; VI-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
; VI-NEXT: flat_load_dword v0, v[0:1]
; VI-NEXT: s_waitcnt vmcnt(0)
; VI-NEXT: v_max_f16_sdwa v1, -v0, -v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
; VI-NEXT: v_max_f16_e64 v0, -v0, -v0
; VI-NEXT: v_or_b32_e32 v2, v0, v1
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_dword v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: v_test_canonicalize_fneg_var_v2f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0
; GFX9-NEXT: v_mov_b32_e32 v1, 0
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_load_dword v0, v0, s[0:1]
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_pk_max_f16 v0, v0, v0 neg_lo:[1,1] neg_hi:[1,1]
; GFX9-NEXT: global_store_dword v1, v0, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: v_test_canonicalize_fneg_var_v2f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s6, 0
; CI-NEXT: s_mov_b32 s7, s3
; CI-NEXT: v_lshlrev_b32_e32 v0, 2, v0
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: s_mov_b64 s[4:5], s[0:1]
; CI-NEXT: v_mov_b32_e32 v1, 0
; CI-NEXT: buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: s_waitcnt vmcnt(0)
; CI-NEXT: v_xor_b32_e32 v0, 0x80008000, v0
; CI-NEXT: v_lshrrev_b32_e32 v1, 16, v0
; CI-NEXT: v_cvt_f32_f16_e32 v1, v1
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
; CI-NEXT: v_mul_f32_e32 v1, 1.0, v1
; CI-NEXT: v_cvt_f16_f32_e32 v1, v1
; CI-NEXT: v_mul_f32_e32 v0, 1.0, v0
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
; CI-NEXT: v_lshlrev_b32_e32 v1, 16, v1
; CI-NEXT: v_or_b32_e32 v0, v0, v1
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%tid = call i32 @llvm.amdgcn.workitem.id.x()
%gep = getelementptr <2 x half>, ptr addrspace(1) %out, i32 %tid
%val = load <2 x half>, ptr addrspace(1) %gep
%fneg.val = fneg <2 x half> %val
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %fneg.val)
store <2 x half> %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @s_test_canonicalize_var_v2f16(ptr addrspace(1) %out, i32 zeroext %val.arg) #1 {
; VI-LABEL: s_test_canonicalize_var_v2f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dword s2, s[0:1], 0x2c
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: s_lshr_b32 s3, s2, 16
; VI-NEXT: v_mov_b32_e32 v1, s3
; VI-NEXT: v_max_f16_e64 v0, s2, s2
; VI-NEXT: v_max_f16_sdwa v1, v1, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v2, v0, v1
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_dword v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: s_test_canonicalize_var_v2f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dword s4, s[0:1], 0x2c
; GFX9-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: v_pk_max_f16 v1, s4, s4
; GFX9-NEXT: global_store_dword v0, v1, s[2:3]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: s_test_canonicalize_var_v2f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dword s2, s[0:1], 0xb
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: s_lshr_b32 s3, s2, 16
; CI-NEXT: v_cvt_f32_f16_e32 v0, s3
; CI-NEXT: v_cvt_f32_f16_e32 v1, s2
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: v_mul_f32_e32 v0, 1.0, v0
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
; CI-NEXT: v_mul_f32_e32 v1, 1.0, v1
; CI-NEXT: v_cvt_f16_f32_e32 v1, v1
; CI-NEXT: v_lshlrev_b32_e32 v0, 16, v0
; CI-NEXT: v_or_b32_e32 v0, v1, v0
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%val = bitcast i32 %val.arg to <2 x half>
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %val)
store <2 x half> %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @test_fold_canonicalize_p0_v2f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: test_fold_canonicalize_p0_v2f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_mov_b32_e32 v2, 0
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_dword v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: test_fold_canonicalize_p0_v2f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_dword v0, v0, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: test_fold_canonicalize_p0_v2f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: v_mov_b32_e32 v0, 0
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> zeroinitializer)
store <2 x half> %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @test_fold_canonicalize_n0_v2f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: test_fold_canonicalize_n0_v2f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_mov_b32_e32 v2, 0x80008000
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_dword v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: test_fold_canonicalize_n0_v2f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: v_mov_b32_e32 v1, 0x80008000
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: test_fold_canonicalize_n0_v2f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: v_mov_b32_e32 v0, 0x80008000
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half -0.0, half -0.0>)
store <2 x half> %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @test_fold_canonicalize_p1_v2f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: test_fold_canonicalize_p1_v2f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_mov_b32_e32 v2, 0x3c003c00
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_dword v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: test_fold_canonicalize_p1_v2f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: v_mov_b32_e32 v1, 0x3c003c00
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: test_fold_canonicalize_p1_v2f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: v_mov_b32_e32 v0, 0x3c003c00
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half 1.0, half 1.0>)
store <2 x half> %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @test_fold_canonicalize_n1_v2f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: test_fold_canonicalize_n1_v2f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_mov_b32_e32 v2, 0xbc00bc00
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_dword v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: test_fold_canonicalize_n1_v2f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: v_mov_b32_e32 v1, 0xbc00bc00
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: test_fold_canonicalize_n1_v2f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: v_mov_b32_e32 v0, 0xbc00bc00
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half -1.0, half -1.0>)
store <2 x half> %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @test_fold_canonicalize_literal_v2f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: test_fold_canonicalize_literal_v2f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_mov_b32_e32 v2, 0x4c004c00
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_dword v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: test_fold_canonicalize_literal_v2f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: v_mov_b32_e32 v1, 0x4c004c00
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: test_fold_canonicalize_literal_v2f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: v_mov_b32_e32 v0, 0x4c004c00
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half 16.0, half 16.0>)
store <2 x half> %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @test_no_denormals_fold_canonicalize_denormal0_v2f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: test_no_denormals_fold_canonicalize_denormal0_v2f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_mov_b32_e32 v2, 0x3ff03ff
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_dword v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: test_no_denormals_fold_canonicalize_denormal0_v2f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: v_mov_b32_e32 v1, 0x3ff03ff
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: test_no_denormals_fold_canonicalize_denormal0_v2f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: v_mov_b32_e32 v0, 0x3ff03ff
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half 0xH03FF, half 0xH03FF>)
store <2 x half> %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @test_denormals_fold_canonicalize_denormal0_v2f16(ptr addrspace(1) %out) #3 {
; VI-LABEL: test_denormals_fold_canonicalize_denormal0_v2f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_mov_b32_e32 v2, 0x3ff03ff
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_dword v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: test_denormals_fold_canonicalize_denormal0_v2f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: v_mov_b32_e32 v1, 0x3ff03ff
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: test_denormals_fold_canonicalize_denormal0_v2f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: v_mov_b32_e32 v0, 0x3ff03ff
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half 0xH03FF, half 0xH03FF>)
store <2 x half> %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @test_no_denormals_fold_canonicalize_denormal1_v2f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: test_no_denormals_fold_canonicalize_denormal1_v2f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_mov_b32_e32 v2, 0x83ff83ff
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_dword v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: test_no_denormals_fold_canonicalize_denormal1_v2f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: v_mov_b32_e32 v1, 0x83ff83ff
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: test_no_denormals_fold_canonicalize_denormal1_v2f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: v_mov_b32_e32 v0, 0x83ff83ff
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half 0xH83FF, half 0xH83FF>)
store <2 x half> %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @test_denormals_fold_canonicalize_denormal1_v2f16(ptr addrspace(1) %out) #3 {
; VI-LABEL: test_denormals_fold_canonicalize_denormal1_v2f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_mov_b32_e32 v2, 0x83ff83ff
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_dword v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: test_denormals_fold_canonicalize_denormal1_v2f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: v_mov_b32_e32 v1, 0x83ff83ff
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: test_denormals_fold_canonicalize_denormal1_v2f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: v_mov_b32_e32 v0, 0x83ff83ff
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half 0xH83FF, half 0xH83FF>)
store <2 x half> %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @test_fold_canonicalize_qnan_v2f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: test_fold_canonicalize_qnan_v2f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_mov_b32_e32 v2, 0x7c007c00
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_dword v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: test_fold_canonicalize_qnan_v2f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: v_mov_b32_e32 v1, 0x7c007c00
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: test_fold_canonicalize_qnan_v2f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: v_mov_b32_e32 v0, 0x7c007c00
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half 0xH7C00, half 0xH7C00>)
store <2 x half> %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @test_fold_canonicalize_qnan_value_neg1_v2f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: test_fold_canonicalize_qnan_value_neg1_v2f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_mov_b32_e32 v2, 0x7e007e00
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_dword v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: test_fold_canonicalize_qnan_value_neg1_v2f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: v_mov_b32_e32 v1, 0x7e007e00
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: test_fold_canonicalize_qnan_value_neg1_v2f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: v_mov_b32_e32 v0, 0x7e007e00
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> bitcast (i32 -1 to <2 x half>))
store <2 x half> %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @test_fold_canonicalize_qnan_value_neg2_v2f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: test_fold_canonicalize_qnan_value_neg2_v2f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_mov_b32_e32 v2, 0x7e007e00
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_dword v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: test_fold_canonicalize_qnan_value_neg2_v2f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: v_mov_b32_e32 v1, 0x7e007e00
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: test_fold_canonicalize_qnan_value_neg2_v2f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: v_mov_b32_e32 v0, 0x7e007e00
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half bitcast (i16 -2 to half), half bitcast (i16 -2 to half)>)
store <2 x half> %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @test_fold_canonicalize_snan0_value_v2f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: test_fold_canonicalize_snan0_value_v2f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_mov_b32_e32 v2, 0x7e007e00
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_dword v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: test_fold_canonicalize_snan0_value_v2f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: v_mov_b32_e32 v1, 0x7e007e00
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: test_fold_canonicalize_snan0_value_v2f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: v_mov_b32_e32 v0, 0x7e007e00
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half 0xH7C01, half 0xH7C01>)
store <2 x half> %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @test_fold_canonicalize_snan1_value_v2f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: test_fold_canonicalize_snan1_value_v2f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_mov_b32_e32 v2, 0x7e007e00
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_dword v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: test_fold_canonicalize_snan1_value_v2f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: v_mov_b32_e32 v1, 0x7e007e00
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: test_fold_canonicalize_snan1_value_v2f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: v_mov_b32_e32 v0, 0x7e007e00
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half 0xH7DFF, half 0xH7DFF>)
store <2 x half> %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @test_fold_canonicalize_snan2_value_v2f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: test_fold_canonicalize_snan2_value_v2f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_mov_b32_e32 v2, 0x7e007e00
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_dword v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: test_fold_canonicalize_snan2_value_v2f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: v_mov_b32_e32 v1, 0x7e007e00
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: test_fold_canonicalize_snan2_value_v2f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: v_mov_b32_e32 v0, 0x7e007e00
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half 0xHFDFF, half 0xHFDFF>)
store <2 x half> %canonicalized, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @test_fold_canonicalize_snan3_value_v2f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: test_fold_canonicalize_snan3_value_v2f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_mov_b32_e32 v2, 0x7e007e00
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_dword v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: test_fold_canonicalize_snan3_value_v2f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: v_mov_b32_e32 v1, 0x7e007e00
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: test_fold_canonicalize_snan3_value_v2f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: v_mov_b32_e32 v0, 0x7e007e00
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> <half 0xHFC01, half 0xHFC01>)
store <2 x half> %canonicalized, ptr addrspace(1) %out
ret void
}
define <3 x half> @v_test_canonicalize_var_v3f16(<3 x half> %val) #1 {
; VI-LABEL: v_test_canonicalize_var_v3f16:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_max_f16_sdwa v2, v0, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
; VI-NEXT: v_max_f16_e32 v0, v0, v0
; VI-NEXT: v_max_f16_e32 v1, v1, v1
; VI-NEXT: v_or_b32_e32 v0, v0, v2
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: v_test_canonicalize_var_v3f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_pk_max_f16 v0, v0, v0
; GFX9-NEXT: v_pk_max_f16 v1, v1, v1
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; CI-LABEL: v_test_canonicalize_var_v3f16:
; CI: ; %bb.0:
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
; CI-NEXT: v_cvt_f16_f32_e32 v1, v1
; CI-NEXT: v_cvt_f16_f32_e32 v2, v2
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
; CI-NEXT: v_cvt_f32_f16_e32 v1, v1
; CI-NEXT: v_cvt_f32_f16_e32 v2, v2
; CI-NEXT: s_setpc_b64 s[30:31]
%canonicalized = call <3 x half> @llvm.canonicalize.v3f16(<3 x half> %val)
ret <3 x half> %canonicalized
}
define <4 x half> @v_test_canonicalize_var_v4f16(<4 x half> %val) #1 {
; VI-LABEL: v_test_canonicalize_var_v4f16:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_max_f16_sdwa v2, v1, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
; VI-NEXT: v_max_f16_sdwa v3, v0, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
; VI-NEXT: v_max_f16_e32 v1, v1, v1
; VI-NEXT: v_max_f16_e32 v0, v0, v0
; VI-NEXT: v_or_b32_e32 v0, v0, v3
; VI-NEXT: v_or_b32_e32 v1, v1, v2
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: v_test_canonicalize_var_v4f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_pk_max_f16 v0, v0, v0
; GFX9-NEXT: v_pk_max_f16 v1, v1, v1
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; CI-LABEL: v_test_canonicalize_var_v4f16:
; CI: ; %bb.0:
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
; CI-NEXT: v_cvt_f16_f32_e32 v1, v1
; CI-NEXT: v_cvt_f16_f32_e32 v2, v2
; CI-NEXT: v_cvt_f16_f32_e32 v3, v3
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
; CI-NEXT: v_cvt_f32_f16_e32 v1, v1
; CI-NEXT: v_cvt_f32_f16_e32 v2, v2
; CI-NEXT: v_cvt_f32_f16_e32 v3, v3
; CI-NEXT: s_setpc_b64 s[30:31]
%canonicalized = call <4 x half> @llvm.canonicalize.v4f16(<4 x half> %val)
ret <4 x half> %canonicalized
}
define amdgpu_kernel void @s_test_canonicalize_undef_v2f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: s_test_canonicalize_undef_v2f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_mov_b32_e32 v2, 0
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_store_dword v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: s_test_canonicalize_undef_v2f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_dword v0, v0, s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: s_test_canonicalize_undef_v2f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: v_mov_b32_e32 v0, 0
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_store_dword v0, off, s[0:3], 0
; CI-NEXT: s_endpgm
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> undef)
store <2 x half> %canonicalized, ptr addrspace(1) %out
ret void
}
define <2 x half> @v_test_canonicalize_reg_undef_v2f16(half %val) #1 {
; VI-LABEL: v_test_canonicalize_reg_undef_v2f16:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_max_f16_e32 v0, v0, v0
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: v_test_canonicalize_reg_undef_v2f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_max_f16_e32 v0, v0, v0
; GFX9-NEXT: v_pack_b32_f16 v0, v0, 0
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; CI-LABEL: v_test_canonicalize_reg_undef_v2f16:
; CI: ; %bb.0:
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
; CI-NEXT: v_mov_b32_e32 v1, 0x7fc00000
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
; CI-NEXT: v_mul_f32_e32 v0, 1.0, v0
; CI-NEXT: s_setpc_b64 s[30:31]
%vec = insertelement <2 x half> undef, half %val, i32 0
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %vec)
ret <2 x half> %canonicalized
}
define <2 x half> @v_test_canonicalize_undef_reg_v2f16(half %val) #1 {
; VI-LABEL: v_test_canonicalize_undef_reg_v2f16:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_max_f16_sdwa v0, v0, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: v_test_canonicalize_undef_reg_v2f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_max_f16_sdwa v0, v0, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; CI-LABEL: v_test_canonicalize_undef_reg_v2f16:
; CI: ; %bb.0:
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
; CI-NEXT: v_mul_f32_e32 v1, 1.0, v0
; CI-NEXT: v_mov_b32_e32 v0, 0x7fc00000
; CI-NEXT: s_setpc_b64 s[30:31]
%vec = insertelement <2 x half> undef, half %val, i32 1
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %vec)
ret <2 x half> %canonicalized
}
define <2 x half> @v_test_canonicalize_undef_lo_imm_hi_v2f16() #1 {
; VI-LABEL: v_test_canonicalize_undef_lo_imm_hi_v2f16:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_bfrev_b32_e32 v0, 60
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: v_test_canonicalize_undef_lo_imm_hi_v2f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_bfrev_b32_e32 v0, 60
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; CI-LABEL: v_test_canonicalize_undef_lo_imm_hi_v2f16:
; CI: ; %bb.0:
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CI-NEXT: v_mov_b32_e32 v0, 0
; CI-NEXT: v_mov_b32_e32 v1, 1.0
; CI-NEXT: s_setpc_b64 s[30:31]
%vec = insertelement <2 x half> undef, half 1.0, i32 1
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %vec)
ret <2 x half> %canonicalized
}
define <2 x half> @v_test_canonicalize_imm_lo_undef_hi_v2f16() #1 {
; VI-LABEL: v_test_canonicalize_imm_lo_undef_hi_v2f16:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, 0x3c00
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: v_test_canonicalize_imm_lo_undef_hi_v2f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_mov_b32_e32 v0, 0x3c00
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; CI-LABEL: v_test_canonicalize_imm_lo_undef_hi_v2f16:
; CI: ; %bb.0:
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CI-NEXT: v_mov_b32_e32 v0, 1.0
; CI-NEXT: v_mov_b32_e32 v1, 0
; CI-NEXT: s_setpc_b64 s[30:31]
%vec = insertelement <2 x half> undef, half 1.0, i32 0
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %vec)
ret <2 x half> %canonicalized
}
define <2 x half> @v_test_canonicalize_undef_lo_k_hi_v2f16() #1 {
; VI-LABEL: v_test_canonicalize_undef_lo_k_hi_v2f16:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_bfrev_b32_e32 v0, 50
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: v_test_canonicalize_undef_lo_k_hi_v2f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_bfrev_b32_e32 v0, 50
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; CI-LABEL: v_test_canonicalize_undef_lo_k_hi_v2f16:
; CI: ; %bb.0:
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CI-NEXT: v_mov_b32_e32 v0, 0
; CI-NEXT: v_mov_b32_e32 v1, 0x41800000
; CI-NEXT: s_setpc_b64 s[30:31]
%vec = insertelement <2 x half> undef, half 16.0, i32 1
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %vec)
ret <2 x half> %canonicalized
}
define <2 x half> @v_test_canonicalize_k_lo_undef_hi_v2f16() #1 {
; VI-LABEL: v_test_canonicalize_k_lo_undef_hi_v2f16:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v0, 0x4c00
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: v_test_canonicalize_k_lo_undef_hi_v2f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_mov_b32_e32 v0, 0x4c00
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; CI-LABEL: v_test_canonicalize_k_lo_undef_hi_v2f16:
; CI: ; %bb.0:
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CI-NEXT: v_mov_b32_e32 v0, 0x41800000
; CI-NEXT: v_mov_b32_e32 v1, 0
; CI-NEXT: s_setpc_b64 s[30:31]
%vec = insertelement <2 x half> undef, half 16.0, i32 0
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %vec)
ret <2 x half> %canonicalized
}
define <2 x half> @v_test_canonicalize_reg_k_v2f16(half %val) #1 {
; VI-LABEL: v_test_canonicalize_reg_k_v2f16:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_max_f16_e32 v0, v0, v0
; VI-NEXT: v_or_b32_e32 v0, 2.0, v0
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: v_test_canonicalize_reg_k_v2f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_max_f16_e32 v0, v0, v0
; GFX9-NEXT: v_pack_b32_f16 v0, v0, 2.0
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; CI-LABEL: v_test_canonicalize_reg_k_v2f16:
; CI: ; %bb.0:
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
; CI-NEXT: v_mov_b32_e32 v1, 2.0
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
; CI-NEXT: s_setpc_b64 s[30:31]
%vec0 = insertelement <2 x half> undef, half %val, i32 0
%vec1 = insertelement <2 x half> %vec0, half 2.0, i32 1
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %vec1)
ret <2 x half> %canonicalized
}
define <2 x half> @v_test_canonicalize_k_reg_v2f16(half %val) #1 {
; VI-LABEL: v_test_canonicalize_k_reg_v2f16:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_max_f16_sdwa v0, v0, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v0, 0x4000, v0
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: v_test_canonicalize_k_reg_v2f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_max_f16_e32 v0, v0, v0
; GFX9-NEXT: v_pack_b32_f16 v0, 2.0, v0
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; CI-LABEL: v_test_canonicalize_k_reg_v2f16:
; CI: ; %bb.0:
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
; CI-NEXT: v_cvt_f32_f16_e32 v1, v0
; CI-NEXT: v_mov_b32_e32 v0, 2.0
; CI-NEXT: s_setpc_b64 s[30:31]
%vec0 = insertelement <2 x half> undef, half 2.0, i32 0
%vec1 = insertelement <2 x half> %vec0, half %val, i32 1
%canonicalized = call <2 x half> @llvm.canonicalize.v2f16(<2 x half> %vec1)
ret <2 x half> %canonicalized
}
define amdgpu_kernel void @s_test_canonicalize_undef_v4f16(ptr addrspace(1) %out) #1 {
; VI-LABEL: s_test_canonicalize_undef_v4f16:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; VI-NEXT: v_mov_b32_e32 v0, 0
; VI-NEXT: v_mov_b32_e32 v1, v0
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: v_mov_b32_e32 v3, s1
; VI-NEXT: v_mov_b32_e32 v2, s0
; VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1]
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: s_test_canonicalize_undef_v4f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: v_mov_b32_e32 v1, v0
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_dwordx2 v0, v[0:1], s[0:1]
; GFX9-NEXT: s_endpgm
;
; CI-LABEL: s_test_canonicalize_undef_v4f16:
; CI: ; %bb.0:
; CI-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT: v_mov_b32_e32 v0, 0
; CI-NEXT: s_mov_b32 s3, 0xf000
; CI-NEXT: s_mov_b32 s2, -1
; CI-NEXT: v_mov_b32_e32 v1, v0
; CI-NEXT: s_waitcnt lgkmcnt(0)
; CI-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0
; CI-NEXT: s_endpgm
%canonicalized = call <4 x half> @llvm.canonicalize.v4f16(<4 x half> undef)
store <4 x half> %canonicalized, ptr addrspace(1) %out
ret void
}
define <4 x half> @v_test_canonicalize_reg_undef_undef_undef_v4f16(half %val) #1 {
; VI-LABEL: v_test_canonicalize_reg_undef_undef_undef_v4f16:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_max_f16_e32 v0, v0, v0
; VI-NEXT: v_or_b32_e32 v0, 0x7e000000, v0
; VI-NEXT: v_mov_b32_e32 v1, 0x7e007e00
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: v_test_canonicalize_reg_undef_undef_undef_v4f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_max_f16_e32 v0, v0, v0
; GFX9-NEXT: v_pack_b32_f16 v0, v0, 0
; GFX9-NEXT: v_mov_b32_e32 v1, 0x7e007e00
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; CI-LABEL: v_test_canonicalize_reg_undef_undef_undef_v4f16:
; CI: ; %bb.0:
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
; CI-NEXT: v_mov_b32_e32 v1, 0x7fc00000
; CI-NEXT: v_mov_b32_e32 v2, 0x7fc00000
; CI-NEXT: v_mov_b32_e32 v3, 0x7fc00000
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
; CI-NEXT: v_mul_f32_e32 v0, 1.0, v0
; CI-NEXT: s_setpc_b64 s[30:31]
%vec = insertelement <4 x half> undef, half %val, i32 0
%canonicalized = call <4 x half> @llvm.canonicalize.v4f16(<4 x half> %vec)
ret <4 x half> %canonicalized
}
define <4 x half> @v_test_canonicalize_reg_reg_undef_undef_v4f16(half %val0, half %val1) #1 {
; VI-LABEL: v_test_canonicalize_reg_reg_undef_undef_v4f16:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_max_f16_e32 v0, v0, v0
; VI-NEXT: v_max_f16_sdwa v1, v1, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_or_b32_e32 v0, v0, v1
; VI-NEXT: v_mov_b32_e32 v1, 0x7e007e00
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: v_test_canonicalize_reg_reg_undef_undef_v4f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: s_mov_b32 s4, 0x5040100
; GFX9-NEXT: v_perm_b32 v0, v1, v0, s4
; GFX9-NEXT: v_pk_max_f16 v0, v0, v0
; GFX9-NEXT: v_mov_b32_e32 v1, 0x7e007e00
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; CI-LABEL: v_test_canonicalize_reg_reg_undef_undef_v4f16:
; CI: ; %bb.0:
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
; CI-NEXT: v_cvt_f16_f32_e32 v1, v1
; CI-NEXT: v_mov_b32_e32 v2, 0x7fc00000
; CI-NEXT: v_mov_b32_e32 v3, 0x7fc00000
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
; CI-NEXT: v_cvt_f32_f16_e32 v1, v1
; CI-NEXT: v_mul_f32_e32 v0, 1.0, v0
; CI-NEXT: v_mul_f32_e32 v1, 1.0, v1
; CI-NEXT: s_setpc_b64 s[30:31]
%vec0 = insertelement <4 x half> undef, half %val0, i32 0
%vec1 = insertelement <4 x half> %vec0, half %val1, i32 1
%canonicalized = call <4 x half> @llvm.canonicalize.v4f16(<4 x half> %vec1)
ret <4 x half> %canonicalized
}
define <4 x half> @v_test_canonicalize_reg_undef_reg_reg_v4f16(half %val0, half %val1, half %val2) #1 {
; VI-LABEL: v_test_canonicalize_reg_undef_reg_reg_v4f16:
; VI: ; %bb.0:
; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; VI-NEXT: v_max_f16_e32 v1, v1, v1
; VI-NEXT: v_max_f16_sdwa v2, v2, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
; VI-NEXT: v_max_f16_e32 v0, v0, v0
; VI-NEXT: v_or_b32_e32 v0, 0x7e000000, v0
; VI-NEXT: v_or_b32_e32 v1, v1, v2
; VI-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: v_test_canonicalize_reg_undef_reg_reg_v4f16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: s_mov_b32 s4, 0x5040100
; GFX9-NEXT: v_perm_b32 v1, v2, v1, s4
; GFX9-NEXT: v_max_f16_e32 v0, v0, v0
; GFX9-NEXT: v_pk_max_f16 v1, v1, v1
; GFX9-NEXT: v_pack_b32_f16 v0, v0, 0
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; CI-LABEL: v_test_canonicalize_reg_undef_reg_reg_v4f16:
; CI: ; %bb.0:
; CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CI-NEXT: v_cvt_f16_f32_e32 v0, v0
; CI-NEXT: v_cvt_f16_f32_e32 v1, v1
; CI-NEXT: v_cvt_f16_f32_e32 v2, v2
; CI-NEXT: v_cvt_f32_f16_e32 v0, v0
; CI-NEXT: v_cvt_f32_f16_e32 v1, v1
; CI-NEXT: v_cvt_f32_f16_e32 v3, v2
; CI-NEXT: v_mul_f32_e32 v0, 1.0, v0
; CI-NEXT: v_mul_f32_e32 v2, 1.0, v1
; CI-NEXT: v_mul_f32_e32 v3, 1.0, v3
; CI-NEXT: v_mov_b32_e32 v1, 0x7fc00000
; CI-NEXT: s_setpc_b64 s[30:31]
%vec0 = insertelement <4 x half> undef, half %val0, i32 0
%vec1 = insertelement <4 x half> %vec0, half %val1, i32 2
%vec2 = insertelement <4 x half> %vec1, half %val2, i32 3
%canonicalized = call <4 x half> @llvm.canonicalize.v4f16(<4 x half> %vec2)
ret <4 x half> %canonicalized
}
attributes #0 = { nounwind readnone }
attributes #1 = { nounwind "denormal-fp-math-f32"="preserve-sign,preserve-sign" }
attributes #2 = { nounwind "denormal-fp-math"="preserve-sign,preserve-sign" }
attributes #3 = { nounwind "denormal-fp-math-f32"="preserve-sign,preserve-sign" }
Reference in New Issue View Git Blame Copy Permalink